Biology subjects

Petrillo, M.

Publications and source records attributed to Petrillo, M..

4 recordsLinked to original sources

bioRxiv2026

Evo 2 as a classification machine: evidence from in-context learning and mechanistic interpretability

In-context learning (iCL) is an emergent capability of Large Language Models (LLMs), allowing them to perform new tasks at inference time using prompt-injected examples. While extensively studied in LLMs, the boundaries of its capabilities and underlying mechanisms remain poorly characterised in genomic Language Models (gLMs). Here, we map the operating regime of Evo 2, a nucleotide-level foundation gLM, across five binary classification tasks spanning biological and artificial sequences. We observe robust iCL on shorter natural sequences (F1=0.902 for miRNA, 0.785 for Toxins), degrading with sequence length, and collapsing at kilobase scale. Strikingly, we find no benefit in model scaling, as the 7B model systematically outperforms the 40B variant. We further find that perplexity, a widely used gLM performance proxy, poorly predicts accuracy. Mechanistic interpretability reconciles these observations: the logit-lens profiling suggests a prediction-generalisation trade-off, while Jacobian Scope indicates models might track prompts' structure rather than the signal-carrying content.

genomics↗

bioRxiv2026

Evo 2's Perception of Single Nucleotide Substitutions in the Genes of Two Plant Model Organisms

Although DNA Large Language Models (DNA-LLMs) offer a path to decoding genetic complexity, our ability to evaluate these models is constrained by our incomplete understanding of the very same genetic syntax and functional logic that these models are trained to learn. In this study we use single nucleotide substitutions that have or have not been observed in living organisms, to evaluate how the DNA-LLM Evo 2 interprets gene sequences from two plant model organisms, Arabidopsis thaliana and Oryza sativa japonica. Using perplexity as a measure of the models confidence, we observe that alleles containing simulated substitutions are perceived, on average, as less likely than those observed in vivo. Although the size of the effect is modest, the effect is statistically significant and robust, suggesting that Evo 2 is aligned with our current understanding of evolutionary selective constraints. This approach is designed to be model-agnostic and species-agnostic and could serve as a generic framework for evaluating the performance of DNA-LLMs.

genomics↗

bioRxiv2022

In Silico Analysis Of The Effects Of Omicron Spike Amino Acid Changes On The Interactions With Human ACE2 Receptor And Structurally Characterized Complexes With Human Antibodies

The new SARS-CoV-2 variant Omicron is characterised, among others, by more than 30 amino acid changes (including 4 deletions and 1 insertion) occurring on the spike glycoprotein. We report a comprehensive analysis of the effects of the Omicron spike amino acid changes in the interaction with human ACE2 receptor or with human antibodies, obtained by analysing the publicly available resolved 3D structures. Our analysis predicts that amino acid changes occurring on amino acids interacting with the ACE2 receptor may increase Omicron transmissibility. The interactions of Omicron spike with human antibodies can be both negatively and positively affected by amino acid changes, with a predicted total loss of interactions only in few complexes. We believe that such an approach can be used to better understand SARS-CoV-2 transmissibility, detectability, and epidemiology, especially when extended to other than spike proteins.

bioinformatics↗

bioRxiv2020

Expanding the Galaxy's reference data

SummaryProperly and effectively managing reference datasets is an important task for many bioinformatics analyses. Refgenie is a reference asset management system that allows to easily organize, retrieve, and share such datasets. Here, we describe the integration of refgenie into the Galaxy platform. Server administrators are able to configure Galaxy to make use of reference datasets made available on a refgenie instance. Additionally, a Galaxy Data Manager tool has been developed to provide a graphical interface to refgenies remote reference retrieval functionality. A large collection of reference datasets has also been made available using the CVMFS repository from GalaxyProject.org, with mirrors across the United States, Canada, Europe, and Australia, enabling easy use outside of Galaxy. Availability and implementationThe ability of Galaxy to use refgenie assets was added to the core Galaxy framework in version 20.05, which is available from https://github.com/galaxyproject/galaxy under the Academic Free License version 3.0. The refgenie Data Manager tool can be installed via the Galaxy ToolShed, with source code managed at https://github.com/BlankenbergLab/galaxy-tools-blankenberg/tree/main/data_managers/data_manager_refgenie_pull and released using an MIT license.

bioinformatics↗